Thin-wall blade laser shock peening deformation control method based on digital twin technology

Abstract

The invention discloses a thin-wall blade laser shock peening deformation control method based on a digital twinning technology, and relates to the technical field of aero-engine blade machining. A simulation agent model is used for achieving rapid simulation of the laser shock peening process of the thin-wall blade, so that a twinborn database is expanded, and according to physical entity information of the thin-wall blade and laser shock peening process parameters and based on data of the twinborn database, a digital twinborn body of the thin-wall blade is constructed through machine learning. By means of a deformation prediction model for laser shock peening of the thin-wall blade in the digital twin, deformation prediction in the machining process of the thin-wall blade is achieved, early warning is provided when the deformation error is too large, feedback control is conducted according to the target deformation requirement, actual machining laser parameters are automatically adjusted, and the deformation control effect is achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of aero-engine blade processing, and particularly relates to a thin-wall blade laser shock peening deformation control method based on digital twinning technology. BACKGROUND

[0002] As a core component of an aero-engine, a blade has an extremely harsh working environment and is required to have high fatigue performance. Laser shock peening technology is one of key means for improving the reliability and prolonging the service life of the blade. However, the blade usually has the characteristics of complex structure, extremely thin wall thickness (usually less than 2 mm) and extremely high profile precision (usually required to be in the micron level). In the laser shock peening process, a huge plasma shock wave pressure can cause plastic deformation of the surface layer material of the blade, so that beneficial residual compressive stress is introduced, and at the same time, uneven stress distribution is formed in the component, causing the blade to deform.

[0003] A traditional deformation control method controls the deformation by adjusting the core parameters of laser shock to control the shock intensity, or by designing a shock strategy, using double-sided alternating shock and symmetric path planning to make the generated residual stress as self-balanced as possible, so as to offset the deformation. However, these methods cannot achieve the effect of real-time monitoring and predicting the deformation, nor do they have the function of feedback adjusting the laser parameters, which can cause the continuous accumulation of deformation errors. SUMMARY

[0004] The purpose of the present application is to provide a thin-wall blade laser shock peening deformation control method based on digital twinning technology, which can monitor and predict the deformation of the deformed thin-wall blade in the laser shock peening process in real time, so as to improve the processing precision.

[0005] To solve the above technical problems, the technical solution adopted by the present application is as follows: A thin-wall blade laser shock peening deformation control method based on digital twinning technology, characterized in that it comprises the following steps: Step 1. Constructing a digital twinning prior database through finite element simulation and experiment: perform thin-wall blade laser shock peening deformation finite element simulation through a simulation software, obtain a large amount of sample data of laser shock peening parameters and thin-wall blade deformation, and verify through an experiment. In the experiment, real-time deformation information of the thin-wall blade is collected by using DIC technology; then, the data generated by the finite element simulation and the experimental data are stored in the database as digital twinning prior data; Step 2. Creating a simulation proxy model according to the digital twinning prior data in the digital twinning prior database; Step 3. Refine the digital twin prior database with the simulation agent model: Through the simulation agent model, the order reduction is accelerated, and the rapid simulation of laser shock peening is realized to obtain a large amount of deformation data under different laser parameters and different blade parameters, which is used to expand the twin data under different working conditions, and to perfect the digital twin prior database, providing comprehensive data information for the training and testing of the subsequent deformation prediction model; Step 4. Construct the digital twin of the thin-walled blade: Based on the physical entity of the thin-walled blade and using the collected multi-modal data and the data of the digital twin prior database, the digital twin of the thin-walled blade is constructed by machine learning; the multi-modal data includes physical entity information, initial shape information, environmental information and laser shock peening process parameters of the thin-walled blade; Step 5. Real-time monitoring of the laser shock peening process of the thin-walled blade through the digital twin, predicting the deformation, and automatically adjusting the actual processing laser parameters according to the target deformation requirements to achieve the deformation control effect.

[0006] Further, in step 1, the simulation software used is ABAQUS simulation software.

[0007] Further, in step 1, the specific method for collecting real-time deformation information of the thin-walled blade by DIC technology is as follows: a high-contrast speckle pattern is applied to the back of the thin-walled blade under laser impact, a high-speed camera is used to capture images of the blade during laser impact, displacement and strain distribution of the thin-walled blade surface after laser shock peening is calculated according to the images, and real-time deformation information of the thin-walled blade is obtained.

[0008] Further, in step 2, the simulation agent model is based on a support vector regression algorithm, and a low-dimensional and efficient simulation agent model is obtained by extracting and training simulation historical data.

[0009] Further, the support vector regression algorithm expression is: ; ; wherein, , represent the upper and lower boundary constraints, is a kernel function, is a bias term, , represents a single laser parameter vector, represents a two-vector Euclidean distance, represents a bandwidth hyperparameter.

[0010] Further, in step 3, the simulation agent model takes the simulated laser power density, spot size, and overlap rate as inputs during the simulation process, and outputs the deformation of the thin-walled blade. The node information and data of the simulation data are extracted and input into the twin space to expand the twin database mapping the relationship between the laser shock peening parameters and the deformation.

[0011] Further, in step 4, the physical entity information of the thin-walled blade includes the type of sheet metal, thickness, and material properties. The environmental information is collected by position sensors and temperature sensors, including the position information and temperature information of the thin-walled blade. The laser shock peening process parameters include laser power density, laser pulse width, impact frequency, scanning path, spot size, spot overlap rate, constraint layer, and absorption layer.

[0012] Further, in step 4, the digital twin includes four modules: a thin-walled blade three-dimensional model construction module, a thin-walled blade processing environment simulation module, a thin-walled blade deformation prediction module, and a thin-walled blade processing visualization module. The thin-walled blade three-dimensional model construction module is used to construct the corresponding thin-walled blade three-dimensional model based on the physical entity of the thin-walled blade, and virtually creates the physical model and assembly relationship of the corresponding object entity. The thin-walled blade processing environment simulation module is used to establish a virtual processing environment and realize the correct motion logic of the virtual entity. Based on the obtained thin-walled blade position data, laser scanning path data, and temperature data of the environment, the thin-walled blade three-dimensional model is simulated by modeling software to realize the correct motion control and processing route of the virtual laser machine tool. The thin-walled blade deformation prediction module is used to establish a laser shock peening deformation prediction model based on real-time collected multi-modal data of the thin-walled blade, and realize real-time prediction of the thin-walled blade deformation through the laser shock peening deformation prediction model. When the predicted thin-walled blade deformation deviates significantly from the target deformation, an early warning can be issued in time. The thin-walled blade processing visualization module is used for users to observe the physical state and various processing information of the thin-walled blade during laser shock peening, to display the deformation cloud map in real time, and to support online adjustment of laser parameters for human-machine collaboration.

[0013] Further, the machine learning algorithm is based on the gradient boosting regression tree algorithm, which combines finite element simulation data, experimental data, and data obtained from the simulation agent model to train and test the deformation prediction model, improve the prediction accuracy and real-time performance, and input real-time collected laser power density, spot size, and overlap rate data into the thin-walled blade deformation prediction module as inputs, and output the deformation value as the prediction model.

[0014] Further, the gradient boosting regression tree algorithm expression is: ; ; In the above formula, is the deformation amount of the tth sample predicted, is the output (weak learner) of the ith tree for the input for fitting the residual, is the input feature vector of the tth sample, containing the key parameters of laser shock peening, is the laser power density, is the pulse width, is the spot diameter, is the number of impacts, is the overlap rate .

[0015] Beneficial technical effects The present application uses machine learning to construct a digital twin of a thin-walled blade based on physical entity information of the thin-walled blade and laser shock peening process parameters and data based on a twin database, uses a deformation prediction model of laser shock peening of the thin-walled blade in the digital twin to realize deformation prediction during the processing of the thin-walled blade and to give a warning when the deformation error is too large, to perform feedback control according to the target deformation requirement, to automatically adjust the actual processing laser parameters, to achieve deformation control effect, and to improve the processing precision. BRIEF DESCRIPTION OF DRAWINGS

[0016] The present application is further described with the help of the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present application, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of the following drawings: Figure 1 is a flowchart of the deformation control method of the present application. DETAILED DESCRIPTION

[0017] In order for those skilled in the art to better understand the technical solutions of the present application, the present application is further described in detail below with the help of the accompanying drawings and specific embodiments, and it should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0018] As Figure 1 shown, the embodiment of the present application provides a thin-walled blade laser shock peening deformation control method based on digital twin technology, comprising the following steps: Step 1. Construct a digital twin priori database through finite element simulation and experiment Specifically, the finite element simulation is established by ABAQUS simulation software, the deformation simulation of the thin-walled blade laser shock peening is carried out by an "explicit-implicit" analysis method, a large number of laser shock peening parameters and thin-walled blade deformation sample data are obtained, and the experiment is verified, finally the data generated by the finite element simulation and the experimental data are stored in the database as the "priori knowledge" of the digital twin.

[0019] The finite element simulation process is as follows: first, a three-dimensional model of the physical entity of the thin-walled blade is established and imported into the ABAQUS simulation software, corresponding material properties are assigned, corresponding constraints are applied, then corresponding laser shock peening parameters are applied according to the planned impact path, the calculation is submitted, the explicit dynamics analysis is completed, and finally the implicit statics analysis is carried out on the basis of the above, and the deformation data of the thin-walled blade after laser shock peening is obtained. In the experiment, the same laser shock peening parameters and blades as in the finite element simulation are used for the experiment, and the DIC technology is used to collect real-time deformation information of the thin-walled blade, so as to verify the whole process of the finite element simulation results, and then obtain high-fidelity simulation and experimental data. The DIC technology is used to coat a high-contrast speckle pattern on the back of the thin-walled blade subjected to laser shock, use a high-speed camera to capture images of the blade during laser shock, use a digital image correlation algorithm to calculate the displacement and strain distribution of the thin-walled blade surface after laser shock peening, and obtain real-time deformation information of the thin-walled blade.

[0020] Step 2. Create a simulation proxy model according to the digital twin priori data in the digital twin priori database; the simulation proxy model is based on a support vector regression algorithm, and a low-dimensional and efficient simulation proxy model is obtained by extracting and training simulation historical data.

[0021] The support vector regression algorithm expression is: ; ; Among them, , respectively represent the upper and lower boundary constraints; is a kernel function, is a bias term for improving the overall fitting; , indicates a single laser parameter vector; indicates a two-vector Euclidean distance, and the smaller the distance, the more similar the "process combination"; indicates a bandwidth hyperparameter for controlling the degree of "local sensitivity".

[0022] Step 3. Improve the digital twin prior database using simulation proxy models: Reduce the order and speed up the simulation by using simulation proxy models to achieve rapid simulation of laser shock reinforcement. This will obtain a large amount of deformation data under different laser parameters and different blade parameters, which will be used to expand the twin data under different working conditions, improve the digital twin prior database, and provide comprehensive data information for the training and testing of subsequent deformation prediction models.

[0023] During the simulation, the simulated laser power density, spot size, and overlap rate are used as inputs to the surrogate model, and the output is the deformation of the thin-walled blade. The node information and data of the simulation data are extracted and input into the twin space to expand the database of the mapping relationship between laser shock strengthening parameters and deformation.

[0024] Step 4. Construct a digital twin of the thin-walled blade. Based on the physical entity of thin-walled blades and utilizing acquired multimodal data and digital twin technology... The data from the verification database is used to construct a digital twin of the thin-walled blade using machine learning; the multimodal data includes the physical entity information, initial shape information, environmental information, and laser shock strengthening process parameters of the thin-walled blade.

[0025] The physical entity information of the thin-walled blade, such as the type of sheet metal, thickness and material properties, is collected, and the initial shape information of the thin-walled blade is acquired using a structured light scanner.

[0026] The laser shock peening process parameters include: laser power density, laser pulse width, number of shocks, scanning path, spot size, spot overlap rate, constraint layer, absorption layer, etc.

[0027] The environmental information of the thin-walled blade includes, for example, motion patterns, machine tool temperature, laser position, thin plate position, clamping method, and strengthening path direction. This environmental information is collected using position sensors, temperature sensors, etc., and all information is preprocessed to obtain a preprocessed dataset.

[0028] The digital twin includes a thin-walled blade 3D model construction module, a thin-walled blade processing environment simulation module, a thin-walled blade deformation prediction module, and a thin-walled blade processing visualization module.

[0029] The thin-walled blade 3D model building module is used to build a corresponding thin-walled blade 3D model based on the physical entity of the thin-walled blade, and to virtually generate the physical model and assembly relationship of the corresponding object entity. The thin-walled blade 3D model is modeled using SolidWorks, and the 3D model is rendered using 3DS Max. Finally, the rendered model is imported into Unity 3D software to obtain a virtual entity that matches the physical entity.

[0030] The thin-walled blade processing environment simulation module is used to establish a virtual processing environment and realize correct motion logic of a virtual entity, and according to obtained thin-walled blade position data, laser scanning path data, temperature data of an environment in which the thin-walled blade is located and the like, an environment simulation is performed on a three-dimensional model of the thin-walled blade by using modeling software, so as to realize correct motion control and a processing route of a virtual laser machine tool; The thin-walled blade deformation prediction module is used to establish a laser shock peening deformation prediction model by using a machine learning algorithm based on real-time collected multi-modal data of the thin-walled blade, and to realize real-time prediction of deformation of the thin-walled blade by using the laser shock peening deformation prediction model, so as to timely give a warning when a predicted deformation of the thin-walled blade deviates from a target deformation too much.

[0031] The machine learning algorithm is based on a gradient boosting regression tree algorithm, and is used to train and test a deformation prediction model by combining finite element simulation data, experimental data and data obtained by using a simulation proxy model, so as to improve prediction accuracy and real-time performance. In an actual processing process, real-time collected data such as laser power density, spot size and overlap rate are taken as inputs of the thin-walled blade deformation prediction module, and a deformation value is taken as an output of the prediction model.

[0032] The gradient boosting regression tree algorithm expression is as follows: ; ; In the above formula, is a predicted deformation amount of a tth sample, is an output (a weak learner) of an ith tree to an input , which is used to fit a residual error, is an input feature vector of the tth sample, which contains key parameters of laser shock peening, is a laser power density, is a pulse width, is a spot diameter, is a number of impacts, is an overlap rate .

[0033] Further, a loss function of the gradient boosting regression tree algorithm is a Huber loss function, which is as follows: ; In the above formula, is a real deformation amount of the tth sample, is a predicted deformation amount of the tth sample, and an absolute value of a subtraction result of the two is recorded as a residual error, is a Huber threshold value, which is used to distinguish a boundary between a "small error" and a "large error".

[0034] Further, the recursive formula of the gradient boosting regression tree algorithm is: ; In the above formula, represents the predicted value of the model before this round of update, represents the total predicted value of the model after update, represents the learning rate, represents the output of the nth regression tree, fitting the current residual.

[0035] The thin-walled blade processing visualization module is used for users to observe the physical state and various processing information of the thin-walled blade in the laser shock peening process, can display the deformation cloud in real time, and supports online regulation and control of laser parameters, and realizes man-machine cooperation.

[0036] Step 5: The laser shock peening process of the thin-walled blade is monitored in real time through the digital twin, the deformation condition is predicted, and a warning is given when the deformation error is too large; and feedback control is carried out according to the target deformation requirement, and the actual processing laser parameter is automatically adjusted to achieve the deformation control effect.

[0037] When the predicted deformation condition of the thin-walled blade in the laser shock peening process deviates from the target shape, the digital twin generates a new set of laser process parameters, which is sent to the controller of the physical processing system in real time through the communication interface, and the physical processing system adjusts the corresponding parameters immediately after receiving the instruction, and executes the subsequent laser shock according to the new setting. After the system executes the new parameters, it will continue to return to the first step and perform real-time monitoring, prediction and decision-making again, forming a continuous closed loop of "perception-analysis-decision-execution", realizing dynamic and adaptive process control.

[0038] The present application constructs a digital twin prior database through finite element simulation and laser shock peening experiment, realizes rapid simulation of the laser shock peening process of the thin-walled blade by using the simulation agent model, thereby expanding the twin database, according to the physical entity information of the thin-walled blade and the laser shock peening process parameters, and based on the data of the twin database, a digital twin of the thin-walled blade is constructed by using machine learning, and a deformation prediction model of the thin-walled blade laser shock peening in the digital twin is used to realize deformation prediction in the thin-walled blade processing process and give a warning when the deformation error is too large, feedback control is carried out according to the target deformation requirement, and the actual processing laser parameter is automatically adjusted to achieve the deformation control effect.

[0039] In addition, the different embodiments or examples described in the specification and the features of the different embodiments or examples can be combined and combined by those skilled in the art without contradiction. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for controlling deformation of laser shock peening of a thin-walled blade based on digital twin technology, characterized in that, Comprise the following steps: Step 1. Constructing a digital twin priori database through finite element simulation and experiment: through the simulation software, the deformation finite element simulation of the thin-walled blade laser shock peening is carried out, a large number of laser shock peening parameters and thin-walled blade deformation sample data are obtained, and the experiment is verified, in the experiment process, the real-time deformation information of the thin-walled blade is collected by using DIC technology; then the data generated by the finite element simulation and the experimental data are stored in the database as the digital twin priori data; Step 2. Create a simulation proxy model according to the digital twin priori data in the digital twin priori database; Step 3. Improve the digital twin priori database by using the simulation proxy model: through the simulation proxy model, the order reduction is carried out to realize the rapid simulation of laser shock peening, so as to obtain a large number of deformation data under different laser parameters and different blade parameters, which is used to expand the twin data under different working conditions, improve the digital twin priori database, and provide comprehensive data information for the training and testing of the subsequent deformation prediction model; Step 4. Constructing the digital twin of the thin-walled blade: based on the physical entity of the thin-walled blade and using the collected multi-modal data and the data of the digital twin priori database, the digital twin of the thin-walled blade is constructed by using machine learning; the multi-modal data includes physical entity information, initial shape information, environment information and laser shock peening process parameters of the thin-walled blade; Step 5. Real-time monitoring of the thin-walled blade laser shock peening process through the digital twin, predicting the deformation, and feedback control according to the target deformation requirement, automatically adjusting the actual processing laser parameters to achieve the deformation control effect.

2. The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 1, characterized in that: In the step 1, the simulation software uses ABAQUS simulation software. 3.The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 1, characterized in that: In the step 1, the specific method of collecting real-time deformation information of the thin-walled blade by using DIC technology is as follows: a high-contrast speckle pattern is coated on the back of the thin-walled blade under laser impact, a high-speed camera is used to capture the image of the blade in the laser impact process, the displacement and strain distribution of the thin-walled blade surface after laser shock peening is calculated according to the image, and the real-time deformation information of the thin-walled blade is obtained. 4.The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 1, characterized in that: In the step 2, the simulation proxy model is based on support vector regression algorithm, and through the extraction and training of simulation history data, a low-dimensional and efficient simulation proxy model is obtained.

5. The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 4, characterized in that, The expression of the support vector regression algorithm is: ； ； wherein, , respectively represent upper and lower boundary constraints, is a kernel function, is a bias term, , denotes a single laser parameter vector, denotes a two-vector Euclidean distance, denotes a bandwidth hyperparameter. 6.The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 1, characterized in that: In the step 3, in the simulation process, the simulated laser power density, spot size and overlap rate and other data are used as the input of the proxy model, and the output is the deformation of the thin-walled blade, and the node information and data of the simulation data are extracted and input into the twin space, so as to expand the twin database of the mapping relationship between the laser shock peening parameters and the deformation. 7.The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 1, characterized in that: In the step 4, the physical entity information of the thin-walled blade includes plate type, thickness and material properties; the environment information is the position information and temperature information of the thin-walled blade collected by using position sensor and temperature sensor; the laser shock peening process parameters include laser power density, laser pulse width, impact times, scanning path, spot size, spot overlap rate, constraint layer and absorption layer.

8. The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 7, characterized in that: In step 4, the digital twin comprises four modules: a thin-walled blade 3D model construction module, a thin-walled blade processing environment simulation module, a thin-walled blade deformation prediction module, and a thin-walled blade processing visualization module. The thin-walled blade 3D model construction module is used to construct a corresponding 3D model of the thin-walled blade based on its physical entity, virtually creating the physical model and assembly relationships of the corresponding object entity. The thin-walled blade processing environment simulation module is used to establish a virtual processing environment and realize the correct motion logic of the virtual entity. Based on the obtained thin-walled blade position data, laser scanning path data, and ambient temperature data, it uses modeling software to visualize the 3D model of the thin-walled blade. The model simulates the environment to achieve correct motion control and processing path of the virtual laser machine tool. The thin-walled blade deformation prediction module is used to establish a laser shock strengthening deformation prediction model based on real-time acquired multimodal data of thin-walled blades using machine learning algorithms. The laser shock strengthening deformation prediction model realizes real-time prediction of thin-walled blade deformation. When the predicted thin-walled blade deformation deviates too much from the target deformation, it can issue an early warning in time. The thin-walled blade processing visualization module allows users to observe the physical state and various processing information of thin-walled blades during the laser shock strengthening process. It can display deformation cloud maps in real time and support online adjustment of laser parameters for human-machine collaboration.

9. The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 8, characterized in that: The machine learning algorithm is based on the gradient boosting regression tree algorithm. It combines data obtained from finite element simulation data, experimental data, and simulation proxy models to train and test the deformation prediction model, thereby improving prediction accuracy and real-time performance. In the actual processing, the real-time collected data such as laser power density, spot size, and overlap rate are used as inputs to the thin-walled blade deformation prediction module, and the deformation value is used as the output of the prediction model.

10. The thin-walled blade laser shock peening deformation control method based on digital twin technology according to claim 9, characterized in that, The gradient boosting regression tree algorithm is expressed as follows: ； ； In the above formula, is the predicted deformation amount of the tth sample, is the output (weak learner) of the ith tree for fitting the residual, is the input feature vector of the tth sample, containing the key parameters of laser shock peening, is the laser power density, is the pulse width, is the spot diameter, is the number of impacts, is the overlap rate .​

Images